Hi Daniel,
thanks for the infos!
The code to register the platform device is in a arch_initcall:
struct platform_device jz_i2c_device = {
.name = "jz_i2c",
.id = 0,
.dev = {
.dma_mask = &jz_i2c_dmamask,
.coherent_dma_mask = 0xffffffff,
},
.num_resources = ARRAY_SIZE(jz_i2c_resources),
.resource = jz_i2c_resources,
};
/* All */
static struct platform_device *jz_platform_devices[] __initdata = {
&jz_usb_ohci_device,
&jz_lcd_device,
&jz_usb_gdt_device,
&jz_mmc_device,
&jz_i2c_device,
};
static int __init jz_platform_init(void)
{
printk(KERN_DEBUG "adding jz platform devices\n");
int ret = platform_add_devices(jz_platform_devices,
ARRAY_SIZE(jz_platform_devices));
printk(KERN_DEBUG "i2c_get_adapter() -> %p\n", i2c_get_adapter(0));
return ret;
}
arch_initcall(jz_platform_init);
The code behind i2c_get_adapter(0) is:
struct i2c_adapter* i2c_get_adapter(int id)
{
struct i2c_adapter *adapter;
mutex_lock(&core_lists);
adapter = (struct i2c_adapter *)idr_find(&i2c_adapter_idr, id);
if (adapter && !try_module_get(adapter->owner))
adapter = NULL;
mutex_unlock(&core_lists);
return adapter;
}
EXPORT_SYMBOL(i2c_get_adapter);
So I think it looks for the id field of the jz_i2c_device which should be 0 but
it is not found.
Am 11.09.2010 um 00:37 schrieb Daniel Glöckner:
> On Fri, Sep 10, 2010 at 09:54:57PM +0200, Dr. H. Nikolaus Schaller wrote:
>> Does anyone have an idea or knowledge how initialization of the
>> i2c subsystem really works in the 2.6.24 kernel?
>>
>> As far as I did understand, the platform initialization file just registers
>> a i2c adapter in some list but does not touch it.
>>
>> The function i2c_get_adapter() should search this list to identify the driver
>> according to some "id" value (which one is this? How is it related to
>> the initialization structure in the platform file?).
>
> - arch_initcall causes jz_platform_init to be executed
> - it registers a platform device with id 0
> - module_init later causes i2c_adap_jz_init to be called
> - it registers a platform driver
> - during registration the list of devices is scanned and all devices
> with the correct name "jz_i2c" are probed
> - this causes i2c_jz_probe to be called
> - it takes the id from the platform_device structure and puts it into
> the adapter's nr field.
> - then the adapter is registered, giving it that number
>
>> Then, the first device that is probed initializes the i2c adapter.
>> But that is far beyond the first call to the old-style i2c_open(), i.e.
>> the adapter is not initialized.
>
> The problem is that module_init is way too late for an i2c adapter.
> Try subsys_initcall instead.
The problem is that there are some hack-style drivers using
i2c_open() where I don't even know when the drivers are initialized.
Well, we could rewrite them to be friendly new-style drivers and
completely get rid of the i2c_open calls, but in a first step I would
like to make them work again to better understand their function.
Do you think I think I can add a call to i2c_jz_probe() to the i2c_open()
wrapper? After I finally manage to find the adapter through i2_get_adapter(0)?
Or do you mean changeing the module_init(i2c_adap_jz_init); to
subsys_initcall(i2c_adap_jz_init); ?
BR,
Nikolaus
_______________________________________________
Mipsbook-devel mailing list
[email protected]
http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/mipsbook-devel
